StorageWriter may lose in-memory operations if unable to initialize a SegmentAggregator.

[andreipaduroiu]

Describe the bug
Each StorageWriter iteration works along these lines

1. Fetch a sequence of operations from DurableLog
2. Process each operation in sequence.
3. When processing them, determine which Segment they belong to and initialize an appropriate SegmentAggregator for that segment if necessary.
4. Flush anything if needed
5. Ack.

The problem is that if, in step 3, the SegmentAggregator.initialize errors out (it needs to do a Storage.getStreamSegmentInfo on LTS), then the entire iteration is aborted and all non-processed operations are lost. They have already been fetched out of the DurableLog and now there's nowhere to get them from again.

Fortunately there are plenty of safeguards and sanity checks in the StorageWriter and SegmentAggregator classes to detect situations like these before any actual data loss or corruption happens (the data itself will still be in Tier 1 since it cannot be truncated out).

To Reproduce
See above.

Expected behavior
StorageWriter should be resilient in the face of SegmentAggregator.initialize errors and not "lose" in-flight operations.